Encode what you’re great at. Hand it to agents one step at a time.
Getting good results out of an agent is like teaching a new junior on the team. You already know how to do that. The work is writing it down.
The Agentic Engineer Journey. What I changed, what broke, and one thing you can paste into your own repo. Every other Monday. Unsubscribe whenever. Your address goes to Buttondown and nowhere else, as described in the privacy policy.
Start here
The three posts most people arrive for, and what each one encodes.
Taming Claude YOLO Mode with Safety Hooks
Encode the safety checks you already run without thinking.
Spec-Driven Development: Planning Templates for AI Agents
Encode how you plan work before any code gets written.
4 Claude Code Primitives: Commands, MCPs, Subagents, Skills
The pieces you encode into, and what each one is good for.
It’s onboarding, not prompting
You wouldn’t hand a new hire a ticket and walk away. You’d tell them the things you stopped noticing you know.
“Run the tests before you push”
a hook that blocks the commit
“Read the spec before you write code”
a planning template
“Never touch prod config”
a permission rule
“We name things like this”
context checked into the repo
“I'll review your first few PRs closely”
evals that gate the merge
None of this is new to you. You’ve just never written it down.
Running this inside a large engineering org? I do that too.
Hire me